AI Plugin Index
Caveman
Plugin by JuliusBrussee · updated · ★ 114k
Makes Claude Code reply in short, caveman-style sentences to cut output tokens. Code, commands and error messages stay exact. Aimed at developers who run long agent sessions. Caveman is published by JuliusBrussee on GitHub.
Install Caveman
Built from the published package, never by AI/plugin marketplace add ahmadhammoud56/MOAA
/plugin install caveman@moaaType both lines inside Claude Code: the first adds the marketplace, the second installs the plugin.
Skills can run scripts on your machine. Some need an API key: check the README.
Last commit 2 days ago
Apache-2.0: permissive, fine for commercial use
Clear description, 114,342 stars, original repository
Downloads a script from the internet and runs it
About Caveman
Caveman is a Claude Code plugin that changes how the assistant writes and reads. Its main part is a short reply style: answers come first, code and commands stay unchanged, and security warnings and irreversible actions get full sentences. A second part shrinks logs, JSON, CSV, YAML and test output before the model reads them. Auto picks a model and effort level for each request and needs a free caveman account. Blocks let the agent save scripts in the repo and rerun them. Install with npx, or add it as a Claude Code plugin. It is for developers who want fewer tokens in coding sessions.
Written from the project's README and checked against it.
What Caveman can do
- Say lessMakes the agent answer in short sentences while code and commands stay exact.
- Read lessCompresses logs, JSON, CSV, YAML and test output before the model reads them.
- AutoChooses the model and effort level for each request. Requires a free account.
- BlocksLets the agent keep scripts it writes in the repo and run them again.
- Waste fixesFinds where tokens are spent and keeps the prompt cache warm while you pause.
- BrowseReturns web pages in a compressed form to the assistant.
- /caveman, /ultracave, /megacaveSkill commands that set the reply style, from standard to shortest.
- caveman status and offShows which parts are on and turns a single part off.
What you need
- Node.js 22.13 or newer for the npx installer
- Claude Pro or Max login, or API keys
- Free caveman account for Auto
- Mac, Linux or Windows
Weekly newsletter
Like Caveman? Get the best new plugins every week
The best new tools, security warnings before you install something risky, and one practical tip. Free, three minute read.
Try asking
- Turn on caveman mode and explain why my React component re-renders.
- Run caveman status and show which parts are on.
- Use Auto to pick the cheapest model that can handle this refactor.
Good to know
- Auto benchmark is only half done and the score gap is too small to call.
- HTML files have no compressor yet and used more tokens in the test.
- Claude web, phone app and Cowork setup was not tested end to end.
- The reply-style token figures were measured on an earlier rule set and not re-run.
Security check
C- Downloads a script from the internet and runs itIn
README.md:curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/v4.0.1/install.sh | bash
Checked automatically: README.md. This looks for known risky patterns and is not a full audit. How grades work
Alternatives to Caveman
All alternatives →
Customer support is a Claude Code plugin from Anthropic for support teams. It helps triage tickets, draft responses, package escalations, and turn resolved issues into knowledge base articles.
Bridges Claude Code to Discord messaging, with built-in access control managed through /discord:access. Listed by anthropics in the Claude Code plugins marketplace for developers who want Discord access managed from their assistant.
iMessage messaging bridge with built-in access control. Reads chat.db directly, sends via AppleScript. Manage pairing, allowlists, and policy via /imessage:access.
Telegram messaging bridge with built-in access control. Manage pairing, allowlists, and policy via /telegram:access.
Transforms codebases into interactive knowledge graphs with visualization, chat, diff analysis, and onboarding guides
Tenant-isolated AI teams that plan work, coordinate tasks, recall approved context, and return evidence-backed results
Caveman: common questions
Does caveman change my prompts?
No. The source says your prompts are never rewritten, because the research it cites found that cavemanning prompts makes answers worse.
Does it work with other agents besides Claude Code?
The source says it also installs for Codex, Gemini CLI, OpenCode and more than 30 other agents through the npx installer.
Are the original files lost when they are compressed?
No. The source says every original stays on your disk and the agent can pull it back at any time.
How do I install Caveman?
Inside Claude Code, add the marketplace with /plugin marketplace add ahmadhammoud56/MOAA, then install the plugin with the second command shown above.
Is Caveman free?
The code is published under the Apache-2.0 licence. The service it connects to may have its own pricing.
Is Caveman safe to use?
It scores 89 of 100. Grades: maintenance A, licence A, quality B. Security check: grade C (downloads a script from the internet and runs it). Review the code before you install. How grades work.
What are alternatives to Caveman?
Similar plugins: Customer support, Discord, Imessage, Telegram.